Predictive CX is a huge part of Spotify’s business model, with its capability to analyze users’ listening habits to create personalized playlists by predicting what songs they might like to hear next. It is also driving the tech giant’s advertising business behind the scenes, using predictive models for forecasting and customer journey mapping. Here we dive into the data science of it all with Ruchika Singh, Spotify’s director of data science and insights.
Singh leads a team of data scientists who ensure that every possible data point that can be used for business monitoring is harnessed to provide insights to Spotify’s stakeholders. These insights can be provided through data science models, dashboards and other research techniques, with the goal of creating well-rounded hypotheses that help drive the business forward.
Singh explains that there is no such thing as a perfect data model. “I have been in this industry for more than a decade, working on data science models, building predictive algorithms, and working with teams that are highly sophisticated at doing this,” she says. “No model is right or wrong, some of them are useful, and some of them aren't.”
Building successful data models
“Really, the secret of building a good data science model is being able to fit the data to the assumptions that you have around the business problem you are trying to solve,” she adds. “You can have a very good predictive model, but if it doesn't move the needle or make the customer happier than they were before, the utility of the model fails.”
For a data model to be useful, Singh says that first it is vital to understand what problem you are trying to solve. “As an example, you're trying to forecast revenue for your business. You can either build a model to predict revenue very accurately, or you can understand the underlying reasons why certain customers spend or don’t spend. While the underlying data set is the same, your goal and your outcomes are completely different.”
“The secret of building a good data science model is being able to fit the data to the assumptions that you have around the business problem you are trying to solve.”
“Secondly, when you have chosen what you are trying to do, you can build different kinds of models. It depends how your metric is behaving, how much bias you have, whether accuracy or scalability are more important. One example is building churn models to predict whether somebody is going to come back and purchase a product with your brand or not. We predict this using machine learning (ML) methods but the underlying question that you're trying to solve is, what is the accuracy we are seeking? Are we trying to accurately predict who is more likely to churn? Or are we trying to predict who is not more likely to churn? Depending on whether you are an insurance company or an e-commerce retailer, your use cases will differ.”
Predictive CX for customer journey mapping
Spotify is built on predictive models, and these need to constantly evaluated and trained. “Predictive modelling or ML-based methods are not a one-time thing,” Singh says. “You build a model, you build more features, you put them into production, and you constantly evaluate how they're doing. It is almost like managing a product of its own, where you have ML-based models that feed into your business decisioning, and you constantly improve the models based on how the market is evolving and how your business is evolving. That is really critical.”
Watch Ruchika Singh discuss Spotify's approach to predictive modelling as part of All Access: Predictive CX 2023
From a CX perspective, these data models need to serve other business processes such as customer journey mapping. As Singh says, if you build a predictive modelling tool, you need to envision who is going to use it and how. “How will this improve the business? How will it impact workflow? Just the way we think of Amazon as an e-commerce product, with personalization built in and all of their models helping to understand what is the next best thing you want to buy. It is because they understand what the customer is doing.”
Creating a culture of data
For organizations to build a culture of data and a strong strategy, Singh says it is important to make allies within the company who will help you progress the narrative around data.
“You might be in an organization where data has always had a critical role, and it's a young company; you know how to deal with data and leaders know how to work with data. Your approach to building a data science team might be completely different compared to a service-oriented, traditional company that hasn't really looked at data as a critical part of their day-to-day operation. Here you're building from the ground up because you need to not just build data products, but you also need to train your people to use data and take decisions using data,” Singh says.
“Having friends and other cross-functional teams that understand the value of what data teams bring to you, and working with them hand-in-hand, really helps you make the most impact and maximize the value of data to the organization.”
“Building a diverse and inclusive team is not just a people management goal, it's actually a goal for bringing better products and better data solutions to teams.”
Cultural diversity is also important here, as Singh explains. “I've learned how you cultivate diversity of skills and thoughts in your teams to make sure that people have different opinions on how to approach a problem, but they also have different opinions from a cultural standpoint. Building a diverse and inclusive team is as important as [how much] it is spoken about, and it's not just a people management goal, it's actually a goal for people to bring better products and better data solutions to teams.
“Often we build teams that are focused so much on impact and business alignment, that we forget that these are data scientists who are trying to build and grow their careers in a space that is highly dynamic. As a data science leader, that is something that you have to keep in the back of your mind; how will this bring value to the business? But also, how will this make the team happy and healthy, because they are your biggest assets.”
- Watch All Access: Predictive CX 2024 on demand